Accounting for Cognitive Aging 1 Running Head: Accounting for Cognitive Aging Accounting for Cognitive Aging: Context Processing, Inhibition or Processing Speed? in Press, Aging, Neuropsychology and Cognition

Age-related deficits in context processing were examined in relationship to two predominant theories of cognitive aging (the Inhibitory Deficit and Processing Speed Models). Older and younger adults completed a measure of context processing (AX-CPT task) as well as a computerized battery of inhibitory tasks: Stroop, garden path sentences, go no-go, and the stopsignal paradigm. Participants also completed a simple processing speed task. After controlling for baseline differences in processing speed, age effects were detected on the AX-CPT. Smaller, but significant age effects were noted on the Stroop and stop-signal tasks, but no significant age effects were found on the garden path sentence and go no-go tasks. Inter-task correlations were weak, providing little evidence for a homogenous or uniform construct of inhibition. The sensitivity of the AX-CPT to cognitive aging is discussed in the context of existing theories of cognitive aging. The authors suggest that deficits in context processing and utilization may be important abilities underlying cognitive aging phenomena. Accounting for Cognitive Aging 3 Accounting for Cognitive Aging: Context Processing, Inhibition, or Processing Speed? The goal of much research on cognitive aging is to identify the core cognitive processes that show age-related changes. In our prior research, we found significant age differences on an AX version of the Continuous Performance Test (AX-CPT; Braver et al., 2001), which we interpreted as reflecting age-related changes in context processing. However, task demands on the AX-CPT are multifactorial, and it is possible that age-related changes in AX-CPT task performance primarily reflect cognitive mechanisms other than context processing. In particular, although the AX-CPT is a task of context processing, successful performance also depends on inhibitory control and processing speed efficiency. The goal of the current study was to determine the extent to which age differences in context processing may be related to, or explained by, age-related changes in either inhibitory function or processing speed. Below, we first review our theory regarding context processing in cognitive aging, and prior experimental work in testing this theory using the AX-CPT. Next, we discuss alternative accounts of cognitive aging: the Inhibitory Deficit (ID) account and the Processing Speed (PS) account. We then describe the current study, in which we test these various accounts in terms of their respective abilities in accounting for age-related changes in task performance. Context Processing Context processing is integral to cognitive control as it allows individuals to internally represent patterns of environmental cues such that these cues can be used to exert control over thoughts and behavioral responses. Context processing involves the formation of an internal representation of context, maintenance of context information over time, and continuous updating of context representations to accurately represent changes in environmental cues (Braver & Cohen, 1999; 2000; 2001; Braver, Cohen, & Barch, 2002; O’Reilly, Braver, & Cohen, Accounting for Cognitive Aging 4 1999). Internal representations of context can be generated from the presentation of a specific prior stimulus, as a result of earlier processing, or from task instructions. Context representations appear to be particularly important to cognitive control in situations where there is strong response competition. As such, context can be helpful for guiding behavior when the appropriate response is infrequent or when a dominant response is no longer appropriate (Barch, Racine, & Braver, in submission). The AX-CPT was specifically designed to examine different aspects of context processing. In the AX-CPT, participants see a continuous stream of single letters, presented in cue-probe pairs. Participants are instructed to make a “target” response when they see the letter “X”, but only if it follows and “A” cue; a “nontarget” response should be made with any other cue-probe pairing. As such, the cue (A or non-A) serves as the context that determines how one should respond to an “X” probe. Seventy percent of AX-CPT trials are target (AX) trials. The remaining 30% are nontarget trials, with 10% “A-Y (Y= any probe other than X), 10% B-X (B= any cue other than A), and 10% B-Y trials. This creates a bias for individuals to respond with a target response to X probes because this is the correct response on the majority of trials. Thus, on BX trials, context must be used to inhibit or override a prepotent response tendency. A second bias is also created by the high frequency of target trials. Healthy individuals will have a bias to make a target response after seeing “A” cues, given that the majority of the time they see an “A,” it is followed by an “X.” Thus, on AY trials context actually causes individuals to false alarm, or respond inappropriately to a probe based on cue information. As such, individuals with intact context representations are likely to demonstrate slowing and elevated rates of error in AY trials relative to BX trials because context representations will hurt AY performance, but aid BX performance. Conversely, individuals with poor context representations are likely to Accounting for Cognitive Aging 5 demonstrate slowing and elevated rates of errors in BX trials relative to AY trials, because poor context representations will not allow them to override prepotent responses in BX trials, and will not cause them to false alarm on AY trials. Empirical support for age differences in context processing between young and old adults comes from Braver et al. (2001), Braver, Satpute, Rush, Racine, & Barch (in press), and Barch et al., (in submission). These previous studies revealed that young adults are able to maintain the context of the A cue, as they demonstrated a characteristic slowing of responses in AY trials relative to BX trials as well as greater errors on AY compared to BX trials. In contrast, older adults did not seem able to use the context of the A cue. The most consistent finding is of disproportionately slower RTs on BX trials compared to younger adults (Braver et al., 2001; Barch et al., in submission), and an absence of the typical slowing of RTs on correct AY trials. In addition, one study also found that older adults showed as many (or more) BX than AY errors (Barch et al., in submission). These results suggest that older adults have a subtle, but impaired ability to engage inhibitory mechanisms particularly when they are required to endogenously maintain the context for appropriate behavioral responding. In addition, Braver et al (2001) found that when context maintenance demands are further raised (by introducing interference information between cue and probe information) older adults demonstrate reliable overt errors in responding as well as consistent patterns of slowing. Thus, characteristic age-related patterns of performance on the AX-CPT for AY and BX trials may surface in response slowing and/or overt response errors, depending on the relative demand for context representation, processing, and maintenance in the task situation. Nevertheless, there are at least two possible alternative interpretations of the data regarding age-related performance changes in AX-CPT performance that are based on existing theories of cognitive aging—the ID and PS models. Accounting for Cognitive Aging 6 The ID Account It is important to consider whether an ID account of cognitive aging can explain performance differences between young and old adults on the AX-CPT. Hasher and Zacks (1988) proposed that many age-related deficits observed across cognitive domains (including selective attention, language, and episodic memory) can be explained by a single common mechanism--declining efficiency in inhibitory function with increased age. According to this model of cognitive aging, older adults (as compared to younger cohorts) cannot appropriately filter incoming information, cannot efficiently and accurately delete irrelevant information from cognitive representations of current task demands, and cannot accurately restrain pre-potent tendencies under changing contextual contingencies. Given these general inhibitory deficits, older adults can appear more challenged across cognitive tasks relative to younger adults. A substantial amount of empirical support for the ID model has amassed, coming from a variety of cognitive tasks including directed forgetting paradigms (Zacks, Radvansky, & Hasher, 1996), the garden path sentence paradigm (Hartman & Hasher, 1991), and Stroop tasks (Daigenault, Braun, & Whitaker, 1992). Nevertheless, several studies have suggested that the relation between age and inhibition may not be so clear (e.g., Connelly & Hasher, 1993; Sullivan & Faust, 1993; Sullivan, Faust, & Balota, 1995). Further, in a large study comparing age-related changes on a variety of inhibitory tasks, including negative priming, response compatibility, the stop-signal paradigm, and the Wisconsin Card Sort Test, Kramer and colleagues (1994) showed that agerelated inhibitory deficits were only present on some of the tested tasks. As previously described, the AX-CPT involves inhibitory control because individuals must use context information to override prepotent response tendencies. On the basis of agerelated deficits in inhibition, the ID model may predict that old adults (as compared to young Accounting for Cognitive Aging 7 adults) show elevated rates of error and disproportionate slowing on BX trials since this trial type requires individuals to overcome prepotent response tendencies. Data from previous studies on the AX-CPT and aging reveal this pattern of deficit on BX trials for old adults, but also demonstrate that old adults show relative improvements in performance (compared to young adults) on AY trials. Although the ID model can account for the pattern of age effects on BX trials, it cannot account for a counterintuitive improvement in performance (relative to young adults) on AY trials. Nonetheless, BX trials do appear to tap into the inhibitory construct that Hasher and Zacks posit to be impaired in older adults. Thus, one useful way to examine this issue further would be to compare performance on the AXCPT to performances on other measures of inhibitory function to better understand the constructs represented by the AX-CPT. As such, one of the goals of the current study will be to compare AX-CPT performance to performance on a range of inhibitory control tasks. The PS Account The AX-CPT is a task that requires speeded responses. For this reason, it is also important to consider whether a PS account of cognitive aging could also account for performance differences between young and old adults on the AX-CPT. Salthouse (1996) proposed a uniquely parsimonious explanation for age-related deficits in cognition. He suggested that age differences in simple processing speed decrease general information processing efficiency across a variety of cognitive tasks including those of attention, working memory, and episodic memory; the decreased efficiency results in age-related deficits. Several studies support this model (e.g., Cerella, 1990; Cerella, 1991; Cerella & Hale, 1994; Cerella, Rybash, Hoyer, & Commons, 1993; Sliwinski & Buschke, 1999). In addition, there is now evidence that the effects of generalized slowing with increased age are more pronounced as task Accounting for Cognitive Aging 8 complexity increases (Cerella, 1990; Cerella, Poon, & Williams, 1980; Hale, Lima, & Myerson, 1991; Salthouse, 1995). Although this evidence suggests that age differences in processing speed might account for age-related deficits in cognitive functions such as context processing, other studies have suggested that significant age-related variance in cognition remains after adequately controlling for baseline differences in processing speed (e.g., Keys & White, 2000). In previous studies of age differences on the AX-CPT (Braver et al., 2001; Braver et al., in press), age differences on the AX-CPT remained significant after controlling for baseline processing speed differences, suggesting that the PS model of cognitive aging does not adequately account for age-related deficits on the AX-CPT. However, these previous studies did not include a separate measure of simple processing speed that could be used to prospectively address the influence of processing speed on AX-CPT task performance. Overview of the Current Study The current study was designed to replicate previous findings of age differences on the AX-CPT and to further investigate the extent to which age effects on the AX-CPT relate to the ID and PS models of cognitive aging. In order to examine this issue, we recruited a large sample of healthy older and younger adults to complete a test battery that included the AX-CPT as well as other commonly used tasks of inhibitory control (the Stroop task, go no-go paradigm, garden path sentence task, and the stop-signal paradigm). The following a priori predictions were made: 1. If age differences in AX-CPT performance are simply due to age-related deficits in inhibitory control, then older adults should show elevated error rates and/or disproportionate slowing in BX trials as this condition requires inhibitory control. In addition, age-related effects on BX trials should be highly correlated with age-related Stroop effects, and the ability to withhold prepotent responses on the garden path sentence, go no-go, and stop-signal tasks. A high Accounting for Cognitive Aging 9 degree of intercorrelation between tasks would suggest convergent validity for an age-related inefficiency on a common ability construct (such as inhibition) across tasks. 2. If age differences in AX-CPT performance are due to baseline differences in processing speed, then age-effects on BX and/or AY trials of the AX-CPT should no longer be significant after statistically controlling for simple motor speed. 3. If age differences in AX-CPT performance are uniquely due to age-related deficits in context processing, then older adults should not only show elevated errors and disproportionate slowing in BX trials, but also demonstrate a relative improvement in performance on AY trials, as revealed by a decreased rate of errors and/or decreased slowing on AY trials relative to young adults. In addition, older adult performance on AY and BX trials should be weakly correlated with performances on the Stroop, garden path sentence, go no-go, and stop-signal paradigms, indicating that the context processing construct is not identical to that of inhibition. Method Participants Fifty-one healthy younger adults (35 women and 16 men, mean age = 19.8 years) and 56 healthy older adults (39 women and 17 men, mean age = 74.8 years) with no history of neurological compromise were recruited. This sample comprises a subset of participants recruited for a larger study on cognitive aging. AX-CPT data from the larger study, which includes data from the sub-sample reported here, has been published in Braver et al. (in press). Participants in the young group were recruited from the Washington University community; older participants were recruited from the older adult volunteer pool in Washington University’s Department of Psychology. Informed consent was obtained from all individuals prior to Accounting for Cognitive Aging 10 beginning participation in the investigation following guidelines set forth by the Washington University Standing Committee on the Use of Human Subjects. Participants were told that they could withdraw consent and discontinue the study at any time. Participants were offered $15 remuneration for their participation. Individuals with medical disorders, neurological disorders, psychiatric disorders, or medication histories that could contribute to cognitive dysfunction were excluded from the study. All participants were asked health status screening questions during initial telephone contact. Individuals with a positive history of neurological disorder, cerebrovascular accident, head injury, learning disability and/or recent drug use were not included in this study. All older adult individuals were administered the Blessed Orientation-Memory-Concentration (BOMC; Katzman et al., 1983) over the telephone in addition to the health status questions. Individuals obtaining five or more errors were not included in this study. Among older adult participants included in this study, the mean BOMC score was 1.14 (SD = 1.33). All participants were administered the Vocabulary subtest from the Wechsler Adult Intelligence Scale—Third Edition (WAIS-III; Wechsler, 1997) to provide an estimate of general intellectual function in the two age groups. In addition, because depression may affect psychomotor speed and general cognitive performance (White et al., 1997), all participants were administered the Beck Depression Inventory (BDI; Beck et al., 1961). Individuals with a score of 10 or higher on the BDI were excluded from all analyses. Materials AX Version of the Continuous Performance Test (AX-CPT). Materials for the AX-CPT are those developed by Braver, Barch, and Cohen (1999). In this paradigm, participants are instructed to make a target response when provided the sequence of an A, immediately followed Accounting for Cognitive Aging 11 by the letter X. In all other cases, participants are instructed to make a nontarget response. The paradigm provides a sequence of letters that cue a target response on 70% of the trials (AX trials), priming the cognitive and motor systems to be predisposed towards a target response. The remaining 30% of trials provide cues to make non-target responses (10% AY trials, 10% BX trials, 10% BY trials). Inhibitory control is examined by examining error rates in AY and BX conditions. Specifically, this paradigm examines how individuals’ respond when provided with an “A” cue followed by a non-X probe (e.g., AY trials) and how they will respond when provided with a non-A cue followed by an X probe (e.g., BX trials). Visual stimuli were a series of letters presented in the middle of a computer-controlled display in 48 point Geneva font subtending a visual angle of approximately 3 degrees. One hundred trials were presented for each participant. There was a 5000-ms delay between cue and probe and a 1000-ms intertrial interval between the probe and the next cue. Dependent variables for this task were median numbers of errors and reaction times for the trials of primary interest (e.g., AY and BX trials). Stroop. Participants were required to name the ink color of a printed word as quickly as possible. Because word reading is automatic, inhibitory control is though to be required to override this prepotent cognitive processing. Incongruent trials were composed of color word names (e.g., red, blue, green) that were presented in a color different from the color associated with the word name (e.g., the word red presented in blue ink). Congruent trials were composed of word names in which the word names is a color different from the ink color in which the word is printed. In congruent trials, the word and ink color are the same. Neutral trials consisted of non-color word names within a single semantic category (e.g., animals: monkey, tiger, bear, cat). Materials for this task and task procedure were similar to those used by Barch, Carter, Hachten, Usher, and Cohen (1999) and consisted of 90 trials: 30 congruent trials; 30 Accounting for Cognitive Aging 12 incongruent trials; and 30 neutral trials. Voice reaction times were recorded via a voice key connected to the computer. Sensitivity of the voice key to participants’ vocal responses was adjusted prior to the onset of task administration. Participants’ responses were entered into a keypad on the computer during task administration in order to record accuracy. The inhibitory effects of the paradigm were examined by comparing performance in the incongruent condition to the performance in the congruent condition (Spieler, Balota, & Faust, 1996). This form of Stroop analysis is typically referred to as the “total” Stroop effect (inhibition minus facilitation). The total Stroop effect may be more likely to reveal inhibitory deficits due to the fact a failure to inhibit attention to the word can lead to slow RTs on incongruent trials, but could actually speed performance on congruent trials, leading to a more pronounced difference in performance between the incongruent and congruent conditions (Barch et al., 1999). The dependent variables for this task were median errors and RTs in the congruent and incongruent conditions. For regression analyses of age effects, we formed two residual scores—one for errors and one for RT. Each residual score reflected performance in the incongruent condition adjusted for an individual’s performance in the congruent condition. In other words, residual scores captured the difference between predicted and observed performance in the incongruent condition. Residuals were used instead of difference scores to take into account baseline differences in speed of processing. Garden path sentences. The garden path sentence task is a measure of implicit memory that examines the ability to abandon prepotent or automatic responses when instructed to do so. The garden path sentence task was developed as a task that taps into the ability to inhibit information from memory. In contrast to other commonly used inhibitory tasks (i.e., Stroop, stop-signal), the garden path sentence task does not require a speeded response. As such, it is Accounting for Cognitive Aging 13 thought to represent a more pure measure of inhibitory function in cognition. Materials for the sentence completion task and task procedure are those developed by Hartman and Hasher (1991). The learning phase consisted of presentation of 28 sentences with highly predictable endings (approximate cloze values = .85; e.g., “Before you go to bed, turn off the light”). For each of these sentences a low probability ending (e.g., stove for this example) is also available. For half of the 28 sentence frames presented, participants were instructed to “abandon” the own high probability ending in favor of the low probability ending; participant-generated endings were confirmed for the remaining 14 sentences confirmed the participant-generated endings for the remaining 14 sentences. After a 5-minute delay, participants were given a “memory” test. They were instructed to read 56 moderate-cloze sentence frames (approximate cloze values = .50) as they appeared on the computer screen and to complete the sentences with the first word that came to mind that made sense. Target word completions were defined as responses from the learning phase that participants were instructed to abandon. Control word completions were defined as responses to sentence frames that were never presented during the learning phase. The dependent variable of interest in the garden path task was calculated from target and control word completions. The primary variable of interest was a residual score reflecting performance in the target condition adjusted for an individual’s performance in the control condition. As such, the residual score created captured the difference between observed and predicted performance in the target condition. That is, target word completions were predicted on the basis of one’s ability to produce control word completions. A residual score was used instead of a difference score to take into account baseline differences in sentence completion. Go no-go. The go no-go is a classic task of inhibitory function that is widely used in the clinical and cognitive neuroscience literatures, including a recent brain imaging study of older Accounting for Cognitive Aging 14 adults (Nielson, Garavan, Langenecker, Stein, & Rao, 2000). The task indexes the ability to suppress responding to a low-frequency target stimulus. Materials for this go no-go task and procedure employed here were similar to those used by Casey et al. (1997). Participants were presented with a sequence of single stimuli (letters) in size 48 font one at a time at the center of the computer screen. Participants were instructed to respond by pressing a button with the index finger of their dominant hand to any sequentially presented stimulus except the number 5. Stimuli were presented for 500 ms with an interstimulus interval of 1000 ms. The number 5 occurred in 25% of the trials. The dependent variable for this task was the number of errors in withholding a respond to the 5 stimulus. Stop-signal. The stop-signal task is similar to the go no-go in that it probes the ability to intermittently suppress responding when instructed. A critical difference is that stop-signal onset times are manipulated in order to estimate the exact time required to stop a given response once it had been initiated. The materials and procedure for the stop-signal task used here were similar to those employed by Williams, Ponesse, Schachar, Logan, & Tannock (1999). The go task was a simple reaction time task that had the same stimulus and presentation parameters as the simple processing speed task described in the following section. Participants were told that occasionally an auditory tone (stop-signal) would occur; when they heard the tone, they were not to respond on that trial (e.g., withhold key press). The stop-signal was a 10-ms 1000-Hz tone generated by the computer. Participants were told that the stops-signal would occur at different times on each trial and that they should not wait for the stop-signal because it would occur randomly and infrequently. The 256 trials were divided into 8 blocks of 32 trials each. The stop-signal was presented randomly on 25% of the trials in each block. Stop-signal trials were presented in a different random order for each block of the task. Participants initiated the onset Accounting for Cognitive Aging 15 of each trial by pressing the space bar to begin. Participants received a block of 12 practice trials (including 4 stop trials) to ensure they understood the task instructions. The stop-signal task was administered using an adaptive procedure that adjusted stopsignal delays in a fashion to achieve a target stop-signal error rate of 50%. At this level of performance, the average stop-signal delay provides an index of inhibitory control. The following procedure was used. Stop-signal delay was initially set at 125 ms (i.e., the presentation of the auditory tone occurred 125 ms following the onset of the go stimulus). If the participant successfully inhibited the motor response, the stop-signal delay was increased by 25 ms on the next stop trial, effectively making it harder to inhibit the motor response. Conversely, if the participant failed to inhibit responding, the stop-signal delay was reduced by 25 ms on the next stop trial. During the first two task blocks the stop-signal value fluctuated towards asymptotic values. Consequently, these blocks were excluded from analyses. The median stop-signal delay interval was then calculated for each participant from the 48 stop-signal trials occurring in Blocks 3 through 8. If the tracking procedure functioned effectively the error rate for the trials on which the median stop-signal delay interval was calculated should be 50% (i.e., the person inhibits the motor response on half of the stop-signal trials and fails to inhibit on the other half). Based on 48 trials the 95% confidence interval for an error rate of 50% would be 35% to 65% errors. Only three participants were excluded from analyses due to error rates outside of this confidence interval. For young adults, there was an average of 47% errors; the average was 55% in old adults. The dependent variable used in this task was a residual score. The residual score captured the median stop-signal delay adjusted for an individual’s go-signal reaction time. The Accounting for Cognitive Aging 16 residual was the difference between the predicted and observed median stop-signal delay. A residual score in place of a difference score in order to account for baseline differences in processing speed. Simple motor speed. A simple motor speed task was administered in order to account for age-related differences in performance that are related to motor slowing. This method has been used in previous studies (e.g., Keys & White, 2000) to identify the unique contribution of age to variability in performance on a specified task of interest. Participants were presented with a fixation point (+) of size 48 at the center of a computer screen followed by a black square of size 96 font on a white background. Once the square appeared on the screen, it remained until the participant made a response. Participants used the index finger of their dominant hand to press a designated button on a button box interfaced with PsyScope software as quickly and as accurately as they could when the stimulus appeared. The onset of the next trial was initiated by pressing the spacebar on the computer keyboard. Participants received 5 practice trials in order to ensure comprehension of task instructions. The preparatory interval between the fixation onset and stimulus onset varied randomly between 1000 and 2000 ms in units of 250 ms (1000, 1250, 1500, 1750, 2000 ms) so that the onset time of the stimulus would be difficult to predict. Ten trials with each stimulus onset delay were presented. The dependent variable from this task was median reaction time from 50 trials for each participant. Procedure Participants completed all experimental measures and laboratory tasks in a single twohour testing session. The order of task administration was counterbalanced across participants within each group, with the exception that the simple processing speed task always immediately Accounting for Cognitive Aging 17 preceded the stop-signal task. All testing was conducted in testing rooms specifically designed to provide appropriate lighting and to minimize potential auditory and visual distraction. Participants were informed of their opportunity to request short breaks between tasks at any time during a given testing session. On all tasks, participants were encouraged to respond as quickly as possible without sacrificing accuracy to performance. The tasks were administered on an Apple Macintosh computer using PsyScope software for stimulus presentation and data collection (Cohen, MacWhinney, Flatt, & Provost, 1993). For all tasks but the Stroop, participants responded to computer tasks by pressing response buttons located on a specifically constructed box connected to the computer, which recorded both response choice and reaction time with 1 ms accuracy. Responses to the targets were made with the index finger of the dominant hand. Responses to the nontargets on the AX-CPT task were made with the adjacent middle finger of the dominant hand. Results An alpha level of .05 was used for all statistical tests. Means and standard deviations for age, education, WAIS Vocabulary, and the BDI are presented in Table 1. Younger adults were slightly less educated than older adult participants, t (105) = 2.81, p < .01; younger adult participants had slightly higher WAIS Vocabulary scores than older participants, t (105) = 3.05, p < .01; and younger adults had slightly lower BDI scores than older adult participants, t (105) = 3.85, p < .01. It is important to note that, despite slight group differences on the BDI, both groups obtained scores well within the “not depressed” range of the normative sample (i.e., a score of 10 or higher is considered mildly depressed). Table 2 includes means, standard deviations, and independent t-test results for all dependent measures included in the study. T-tests are provided in order to document age Accounting for Cognitive Aging 18 differences on the various tasks administered. Arcsine transformations were performed on all raw accuracy data before these data were subjected to statistical analysis. T-test results reveal that younger adults made more errors on AY trials than older adults. Compared to young adults, old adults were generally slower across all tasks and they produced greater errors in withholding a prepotent response on the stop signal task. In order to test specific a priori theoretical hypotheses regarding differential aging effects across task conditions, multiple analysis of variance (MANOVA) and hierarchical regression procedures were conducted; the results of finer grained analyses are presented for each task. A/X version of the Continuous Performance Test (A/X CPT). A priori hypotheses were tested using a mixed model analysis of variance (ANOVA) for errors, with age group (young vs. older) as a between-subject independent variable and trial type (AY, BX) as a within-subject independent variable. Figures 1A and 1B reveal AX-CPT error and reaction time data across trial types and facilitate comparing the results of the current study with previously published data on cognitive aging and AX-CPT performance. Analyses of error rates revealed that younger adults produced more errors than older adults, F (1, 105) = 12.29, p < .05, and, participants made more AY than BX errors, F (1, 105) = 5.34, p < .05. Finally, a significant interaction between age group and trial type was found, F (1, 105) = 7.77, p < .05. Consistent with prediction, simple effects analyses revealed that younger adults produced more AY errors than BX errors, F (1, 105) = 12.42, p < .0001. Thus, the context of an A cue appeared to cause young adults to false alarm on trials that should have resulted in a nontarget response. In contrast, simple effects analyses revealed that older adults produced approximately equivalent amounts of errors in AY and BX conditions, F (1, 105) = .12, p > .05, further, the age group X trial type interaction was also driven at least in part by the fact that younger adults Accounting for Cognitive Aging 19 produced substantially more AY errors than older adults F (1, 105) = 20.47, p < .0001, whereas, there was no significant group difference in BX errors, F (1, 105) = .79, p > .05. These results suggest that as predicted, the presence of context resulted in greater errors for younger than older adults, whereas older adult performance appeared less dependent on the context of the cue (A or non-A) provided. However, we did not find the predicted increase in BX errors among older adults. RT analyses revealed that older adults were generally slower than younger adults, F (1, 105) = 55.48, p < .01. A main effect of trial type was found, F (1, 105) = 62.25, p < .01, as well as a group x trial type interaction, F (1, 105) = 24.04, p < .01. Simple effects analyses revealed that young, F (1, 105) = 78.18, p < .0001, and older adults, F (1, 105) = 4.68, p < .05, were slower in AY relative to BX trials. Thus, both groups demonstrated RT slowing on trials during which they received the A cue (context information). Older adults were slower than younger adults on both AY trials, F (1, 105) = 33.45, p < .0001, and BX trials, F (1, 105) = 51.08, p < .0001. However, comparison of the effect sizes for group differences on these two trial types indicated that age differences in RT were disproportionately greater on BX trials relative to AY trials. Although old adults did not demonstrate the predicted elevation of overt errors on BX trials, RT data provide evidence that older adults were less able to use the context provided by the cue to override pre-potent response tendencies. In order to determine that confirmed age effects on the AX-CPT were not simply due to motor slowing, hierarchical regressions were conducted with statistical control for SRT. After controlling for motor slowing, the unique effect of age on AY error performance, R = .14, β = .46, F (1, 104) = 17.77, p < .01, and BX RT performance remained significant, R = .04, β = .23, F (1, 104) = 7.05, p < .01. Interesting, the effect of age on AY RT was no longer significant after Accounting for Cognitive Aging 20 controlling for simple processing speed, suggesting that the initially detected relationship was likely related to age differences in cognitive efficiency. Stroop. Hierarchical regression was used to examine age differences on the total Stroop effect. Performance in the incongruent condition was the dependent variable. Performance in the congruent condition was entered as the independent variable on the first step; age was entered at the second step of the analyses. After entering error performance in the congruent condition first, the effect of age on error performance in the incongruent condition was not significant, R = .02, β = .13, F (1, 104) = 1.88, p > .05. After entering RT performance in the congruent condition first, the effect of age on RT performance in the incongruent condition was significant, R = .03, β = .19, F (1, 104) = 13.97, p < .05. Therefore, the Stroop effect increased with increased age. Even after accounting for motor slowing first (with the SRT task), the Stroop effect increased with increased age, R = .01, β = .15, F (1, 103) = 7.09, p < .05. Garden path sentences. Table 2 shows no significant group differences on any garden path variables. Further analyses with hierarchical regression were conducted to examine the unique relation between age and target completion performance over and beyond the ability to provide control completions. Target completions were prorated based on a person's tendency to produce anticipated critical endings during the learning phase (e.g., number of critical items from the learning phase that a participant actually generated and were disconfirmed by the experimenter that were expected to be generated as high-cloze endings). In the hierarchical regression the prorated number of target completions was entered as the dependent variable. The total number of control completions was entered as the independent variable in the first step of the analysis; age was entered as the independent variable in the second step. After entering control completions first, the effect of age on critical completions was not significant, R = .00, F Accounting for Cognitive Aging 21 (1, 104) = 0.46, p > .05; increased age was not associated with a greater recall of target items during the test phase. Additional analyses indicated that there was no age difference in raw target completion performance (i.e., performance before prorating) during the learning phase of this task, t (105) = 1.10, p > .05. Go No-Go. No significant group differences were found on any of the go no-go dependent variables when evaluated by t-tests, or hierarchical regression. There were no significant effects of age on the percentage of errors in the no-go condition, R = .00, β = .-.02, F (1, 105) = 0.06, p > .05, or in the go condition, R = .03, β = .17, F (1, 105) = 3.03, p > .05. Stop-Signal. Hierarchical regression was used to determine if there were age effects in the mean time needed to respond accurately to the stop signal. Mean stop-signal delay was entered as the dependent variable. Median go-signal reaction time was entered as the independent variable on the first step of the analysis; age was entered as the independent variable on the second step. As predicted, stop signal reaction time increased with age, R = .01, β = -.14, F (1, 104) = 11.31, p < .01. Despite age-related slowing effects, increased age appears to result in the need for a decreased stop-signal delay, R = .01, β = -.13, F (1, 103) = 8.27, p < .01. Simple motor speed. A significant positive correlation between age and reaction time was confirmed, r = .57, p < .01, providing evidence for motor slowing with increased age. Unilateral ANOVA confirmed significant group differences in speed, F (1, 107) = 50.03, p < .0001 (partial eta-squared = .36). Comparison of cognitive aging effects across tasks. The magnitude of cognitive aging effects across tasks, after controlling for processing speed, is summarized and presented in Table 3. As shown, the magnitude of the beta weights suggests that the largest age-related effects were found in the two AX-CPT measures. This numerical effect was more directly examined using Accounting for Cognitive Aging 22 the Z-test of correlated correlation coefficients (Meng, Rosenthal, & Rubin, 1992). This approach takes into account the degree to which performance on one task may be correlated with performance on a second task. Z-test results revealed that the unique age effect yielded by AY errors on the AX-CPT was significantly greater than unique age effects yielded by BX errors on the AX-CPT, Z (107) = 3.32, p < .01, and by Stroop reaction times, Z (107) = 3.32, p < .01. The unique age effect yielded by AY errors on the AX-CPT was not significantly greater than age effects yielded on the stop signal task, Z (107) = .70, p > .01. The unique age effect yielded by BX RT on the AX-CPT was not significantly greater than age effects yielded by the Stroop, Z (107) = .10, p > .01, or stop signal tasks, Z (107) = .07, p > .01. Finally, the unique age effected yielded by the Stroop task was not significantly greater than age effects yielded by the stop signal task, Z (107) = .03, p > .01. Correlations between tasks. Correlations were computed in each age group to examine the degree to which performance across tasks was related. In the older adult group, AY errors significantly correlated with BX errors (r = .30, p < .05) and Stroop RT (r = -.32, p < .05). BX RT significantly correlated with BX errors (r = .32, p < .05), AY RT (r = .40, p < .01), and median SRT (r = .56, p < .01). AY RT significantly correlated with SRT (r = .35, p < .01). All other inter-task correlations were not significant. In the younger adult group, go no-go performance was significantly correlated with AY errors (r = .31, p < .05), Stroop errors (r = .44, p < .01), stop signal (r = -.35, p < .05), and SRT (r = -.33, p < .05). BX RT significantly correlated with AY RT (r = .53, p < .01) and SRT (r = .40, p < .01). AY RT significantly correlated with Stroop RT (r = -.36, p < .01). All other inter-task correlations were not